Level Set Evolution Based On Global Fitting And Object Sign Function For Image Segmentation

As an important research direction in the field of image processing,image segmentation is widely used in artificial intelligence,three-dimensional reconstruction,medical image analysis and many other fields.It can detect the region of interest from the background in one image,which is the basis of subsequent target recognition and feature analysis.However,real images are often disturbed by noise,and there are some problems such as blurred boundary and intensity inhomogeneity,which make image segmentation difficult.Among the existing image segmentation methods,the level set evolution model has been widely concerned and applied because of its rigorous mathematical theory,easy algorithm implementation,and the ability to deal with the topological structure that the traditional parametric active contour models cannot handle.There are many level set algorithms.This paper studies the classical distance regularized level set evolution(DRLSE)model,which can effectively segment images with intensity homogeneity.But its initial contour must completely contain the target or be set inside the target.Besides,it has poor anti-noise ability and is easy to leak from weak boundary.In view of these problems,this paper improves this model in some aspects as follows:(1)An object sign function is designed to enhance the robustness of the DRLSE model to initial contotur.By using the difference between image intensity and the average intensity of the region inside and outside the contour,this function can have the opposite signs on both sides of the target boundary,which enables the contour to adaptively choose evolution direction according to this feature,thereby reducing the dependence on the position of initial contour.At the same time,due to the different value adjusted according to image contrast,it improves the ability of capturing boundary.(2)Global fitting information is introduced into the model to improve the segmentation speed.After several iterations of the initial contour with the model combined with two global fitting functions,which are designed based on the threshold of Otsu algorithm,the updated contour will arrive near the boundary quickly,and then use the improved DRLSE model to refine the contour.This process can effectively improve the segmentation efficiency.(3)Aiming at the images with several adjacent targets,region growth matrix is added to the area term of the DRLSE model.The gradient information of the filtered image will become blurred.Therefore,if the targets are too close,the gradient edges will be connected and finally lead to incorrect result.The region growth matrix can enhance the energy between adjacent regions,which helps accelerate the evolution speed,so that the model can overcome the failed segmentation caused by widened gradient.(4)By defining a new adaptive boundary indicator function,which has different value according to different image information at the same gradient,the energy can be adaptively increased or decreased in different regions of different images,such that this model has the capacity of driving the motion of contour at a more appropriate speed.It not only enhances the anti-noise ability,but also reduces the occurrence of boundary leakage.Moreover,it can bring faster segmentation speed.In summary,this paper proposes a new adaptive level set algorithm based on the DRLSE model for image segmentation.The experimental results show that the improved model has good robustness to the initial contour and better segmentation results for images with weak boundary and big noise.In addition,it can effectively capture boundary of adjacent multi-objective images.At the same time,the segmentation speed has also been improved.